Energy demand forecasting with AI is a machine-learning problem, not a chatbot one. We build demand, solar and wind forecasts that report a range, prove themselves against your current method, and leave the decision to the operator.
All Energy & Utilities →
Demand used to be shaped mainly by temperature, daylight and the calendar. Rooftop solar now hides consumption behind the meter on sunny afternoons, heat pumps make winter demand more sensitive to cold snaps, and electric vehicle charging moves load into the evening and the night. Generation from wind and solar depends on weather that is itself forecast. A single-number forecast hides all of that uncertainty from the person who has to act on it.
So we forecast distributions. A day-ahead load forecast comes with quantiles; a solar power forecast says how wide the range is when cloud cover is uncertain. Traders can size positions, and control rooms can plan reserve, around the honest width of the range rather than a point that will be wrong most of the time.
Every model is backtested on held-out history against two references: a naive persistence forecast and the method you use today. If it does not beat both, by a margin that matters to your imbalance costs, we say so and stop. We report accuracy separately for the days that hurt (holidays, heatwaves, cold snaps, sudden cloud) because averages flatter forecasts on the easy days.
In production, forecast error and calibration are monitored daily, and a drift alert goes to a person when the inputs or the errors move outside their recent range. Where forecasts or the data behind them could be inside information under REMIT, access and logging are designed with your compliance team.
It is the use of machine-learning models to predict electricity or gas demand from historical consumption, weather forecasts, the calendar and other drivers. Modern approaches produce a range of likely outcomes rather than a single number, which matters more as rooftop solar, heat pumps and electric vehicles make demand less predictable.
Not for the numbers. Demand and generation forecasting is done best by well-tested statistical and machine-learning models that can be backtested and explained. A language model is useful alongside them, to summarise why a forecast moved for the people on the desk, but it never sets the forecast itself.
By backtesting on held-out history against a persistence baseline and your current method, reported separately for clear, mixed and overcast days. We check both accuracy and calibration, meaning whether the stated ranges contain the outcome as often as they claim. A forecast with honest ranges is more useful to an operator than a sharper one that is overconfident.
What we built, what stage it is at, and a link to the live thing wherever there is one.
yourOS is an operating system with an AI at its heart. You talk to it like a person, out loud or typed. It makes your calls, plans your days and works on your projects with a crew of agents, on your own private server. Our own product, built from the first commit; the waitlist is open.
Website, backend and app assets for Sonora, a free AI sound-therapy app on iOS and Android published by Cognitive AI.
Brand, website, prototype, apps on five platforms, the AI backend and the algorithm-intelligence pipeline for Podswap, a social growth engine that runs without a human in the loop.
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Live at coherence.com: brand identity generated with AI agents we directed, the motion system and website for the group behind Sonora, AboutAI, WakeUp and MudShadows. Agent work remains in build.
Live at domainbooth.com for DomainBooth, a premium domain brokerage: a new animated mark, every page redesigned and rebuilt with designed mobile layouts, and a custom CMS for its sales record, featured names, press and enquiries.
Give us two years of history and your current forecasts. The backtest decides whether there is a project.
30-minute technical call · no deck